I build AI-powered infrastructure for learning, research, and work. Not prompts. Not chatbots. Systems — structured, purposeful, built to help people understand better.
Not conclusions. Evolving convictions — shaped by 3+ years of building, testing, and often getting things wrong first.
A grounded legal AI assistant built around Belarusian law. Structured retrieval, citation grounding, constrained reasoning — designed with explicit failure modes and trust architecture from the ground up.
"The challenge wasn't making it answer. It was designing the system to know when not to — and explain why."
Educational content generation using NotebookLM as core reasoning engine. Transforms dense source material into structured, interactive knowledge — prioritising genuine understanding over output speed.
Reusable prompt workflow infrastructure for content research, drafting, and refinement. Built around structured frameworks — designed to be operated by non-engineers as a repeatable production system.
An evolving collection of reusable Claude skills designed for research, workflow design, learning systems, and practical AI applications. Currently documenting, testing, and publishing skills through open-source experiments and real-world use cases — an active and growing body of work built through craft, not volume.
A methodical cost analysis of the AI video generation landscape. Separating hype from real economics, comparing major platforms on output quality per dollar — what creators actually need to know before committing budget.
A practical framework for educators — a system for using AI in lesson design that prioritises understanding and pedagogical integrity over speed alone.
Why the distinction between AI that produces answers and AI that builds comprehension matters — and what design principles follow from it.
Open to remote opportunities and international teams. I work best with people who care about depth, clarity, and systems that actually work — not just ship.